convez.ai
MCP ServerIntegração & APIsAvançado
TGYD-helige

Pi — Pacotes Compartilhados para Runtime de Agents

TGYD-helige/pi

Arquitetura composável para agents com persistência, automação e extensões

Ver no GitHub
BomPluginScore · de 0 a 100

Conjunto modular de pacotes TypeScript que fornece contratos de runtime, adaptadores de persistência e extensões para aplicações Pi Agent. Inclui suporte a automação web/desktop, agendamento, telemetria, integração com plataformas de comunicação (Feishu, WeCom, DingTalk) e geração de mídia (vídeo, imagem).

Instalação

$npx convez-ai TGYD-helige/pi

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
93
Forks
20
Acompanhando
0
Saúde
50%
Último commit
há 1 dia
Versão
v0.1.40
Issues abertas
1
Contribuidores
21

Casos de uso

  • Construir agents autônomos com persistência de sessão e memória curada
  • Integrar automação web, desktop e agendamento de tarefas em aplicações de IA
  • Implementar telemetria e rastreamento de custos com Langfuse/OpenTelemetry
  • Conectar agents a plataformas de workspace (Feishu, WeCom, DingTalk)

Veredito

MCP server robusto e bem-estruturado para ecossistema Pi. Oferece componentes de qualidade produção (storage, telemetria, automação), mas requer experiência avançada em TypeScript e conhecimento da arquitetura Pi para integração efetiva.

README

A tradução para o português deste README ainda está na fila.

Pi

Pi packages preview

Shared TypeScript packages for Pi runtime applications.

On Pi 0.99+, video, teamwork, curated memory, Mem0, and scheduler tools expose bounded structured results for codemode scripts. Scripts can consume IDs, complete memory entries/task prompts, media paths, and available QC metadata directly; see the corresponding package READMEs for result contracts.

This repository contains the open-source runtime contracts, adapters, and orchestration helpers that are consumed by Pi Agent and related applications. The packages are intentionally small and composable: host applications provide HTTP routing, authentication, model runtime setup, deployment configuration, and product-specific UI.

Packages

CategoryPackagePurpose
Core@amaster.ai/pi-sharedShared runtime types and contracts: settings loader, session/event/artifact types, turn and subagent types.
Core@amaster.ai/pi-storageJSON-file and MySQL/Prisma persistence adapters for sessions, transcripts, events, memory, artifacts, subagents, and scheduled tasks.
Extension@amaster.ai/pi-attachmentsAttachment normalization, local/remote upload handling, document parsing, and model-readable attachment prompts.
Extension@amaster.ai/pi-telemetryRuntime traces, TTFT, token/cost accounting and diagnostics with Langfuse, Arize Phoenix and OpenTelemetry exporters.
Extension@amaster.ai/pi-task-schedulerCron-based scheduled task management with LLM-callable tools.
Extension@amaster.ai/pi-goalDerives a goal from the conversation and keeps the agent working until the condition is met, with iteration/token backstops.
Extension@amaster.ai/pi-browser-useBrowser automation via chrome-devtools-mcp, with opt-in live-session retention and cloud lifecycle adapters.
Extension@amaster.ai/pi-web-accessWeb search (including Cloudflare), URL content extraction, and image search across configurable providers.
Extension@amaster.ai/pi-computer-useCross-platform computer-use tools for desktop automation.
Extension@amaster.ai/pi-channelsNative messaging channels: Feishu, WeCom, and webhooks.
Extension@amaster.ai/pi-memoryPersistent curated memory (MEMORY.md + USER.md) injected into the system prompt as a snapshot frozen between compactions.
Extension@amaster.ai/pi-securityResource-aware security policy engine and tool authorization.
Extension@amaster.ai/pi-teamworkTeam collaboration and issue management via Multica.
Extension@amaster.ai/pi-image-genImage generation via OpenAI gpt-image, Google Nano Banana, Alibaba Qwen-Image, OpenRouter, and custom providers.
Extension@amaster.ai/pi-video-genDirect video-model clips, mixed-shot batch rendering with recovery, and local concat, FFmpeg timeline or editable HyperFrames (default) / Remotion composition. Includes subtitles, QC and bundled FFmpeg runtimes.
Extension@amaster.ai/pi-larkLark/Feishu workspace integration through lark-cli, including calendar, docs, drive, sheets, Base, tasks, mail, wiki, and IM skills.
Extension@amaster.ai/pi-wecomWeCom workspace integration through wecom-cli, including contacts, messages, meetings, schedules, todos, docs, and smart sheets.
Extension@amaster.ai/pi-dingtalkDingTalk workspace integration through dws CLI, including calendar, docs, chat, todos, sheets, AI tables, approvals, mail, wiki, and meeting minutes.

Core packages provide types and persistence used by every host application. Extension packages declare their Pi runtime entry point in the pi.extensions field of package.json, and Pi loads it from there.

Every package is ESM-only and published under the @amaster.ai npm scope.

Extension Previews

Requirements

  • Node.js >=24
  • pnpm 10.18.3

Use Corepack when possible:

corepack enable
corepack install -g pnpm@10.18.3

Development

Install dependencies:

pnpm install

Run the full local check:

pnpm run pr-check

Common commands:

pnpm build
pnpm typecheck
pnpm test
pnpm --filter @amaster.ai/pi-storage prisma:generate

@amaster.ai/pi-storage includes a Prisma schema at packages/storage/prisma/schema.prisma. The root build and typecheck scripts generate the Prisma client before compiling project references.

Consuming Packages

Install only the packages your application needs:

pnpm add @amaster.ai/pi-shared @amaster.ai/pi-storage

Most packages expose a root entry point. Some packages also expose focused subpath entry points:

import { createRuntimeStorage } from "@amaster.ai/pi-storage";
import { JsonRuntimeStorage } from "@amaster.ai/pi-storage/json";
import { loadPiSettings } from "@amaster.ai/pi-shared/settings";
import { createLangfuseExporter } from "@amaster.ai/pi-telemetry/langfuse";
import { createOtelExporter, createPhoenixExporter } from "@amaster.ai/pi-telemetry/otel";
import memoryExtension from "@amaster.ai/pi-memory/extension";

Pi loads an extension package from the entry declared in its package.json pi.extensions field. Host applications that import an extension directly can use the package's default export; pi-goal, pi-memory, pi-security, pi-task-scheduler, and pi-telemetry also expose it as an ./extension subpath.

See each package README for package-specific examples and public API notes.

DeepSeek Harness

Compatibility depends on the host, DSH, Pi, and extension versions. Follow the selected host's documentation for installation and current support.

HostLoading model
pi2dshDiscovers explicitly installed Pi packages in a DSH profile.
dsh-pi-hostLoads package names configured in its extensions list.

License

Apache-2.0

Ver o original no GitHub

Bom para

Relacionados